[Model] Add DeepSeek-V4 CPU backend (#55355)
Signed-off-by: jiang1.li <jiang1.li@intel.com>
This commit is contained in:
@@ -70,12 +70,14 @@ def main(
|
||||
|
||||
packed_w13 = cpu_prepack_moe_weight(w13, isa)
|
||||
packed_w2 = cpu_prepack_moe_weight(w2, isa)
|
||||
output = torch.empty_like(input_tensor)
|
||||
|
||||
def run_benchmark(iters: int) -> list[float]:
|
||||
times = []
|
||||
for _ in range(iters):
|
||||
start_time = time.perf_counter_ns()
|
||||
_ = cpu_fused_moe(
|
||||
cpu_fused_moe(
|
||||
output,
|
||||
input_tensor,
|
||||
packed_w13,
|
||||
packed_w2,
|
||||
|
||||
Reference in New Issue
Block a user